Quiz 2 Answer Explanation


Why is this answer choice wrong? It is true that without skip connections neural networks typically first improve then get worse on training accuracy as the number of layers increases. Therefore, adding more layers may or may not hurt training set performance, which implies that the choice should be correct?

Hi dg2004,

My interpretation is that making an inception network deeper should not hurt training set performance, because the identity function can be learned for each inception block. See this explanation.